LES of a simplified HVAC system used foraero-acoustic predictions

S. Rolfo1

C. Moulinec 1 D. R. Emerson 1

1STFC Daresbury Laboratory, Warrington, Cheshire WA4 4AD, United Kingdomstefano.rolfo@stfc.ac.uk

April 2nd 2015

Rolfo et al. HVAC

Overview

1 Heat and Ventilation Air Conditioning: HVACIntroductionFlow FeaturesAcoustic

2 Paraview in a Visualization Cluster (Placement of E. Harrison)

3 Conclusions and Future Work

Rolfo et al. HVAC

HVAC: Introduction

HVAC

HVAC stands for Heat and VentilationAir Conditioning and it is a commonsystem in several engineeringapplication

This simplified configuration has beenproposed by German car industry forvalidation of numerical methods

The system is a composed by a ductbend with a flap

The duct creates a jet in an openspace

Both experimental data for fluid flow(PIV) and aeroacoustic are available

Rolfo et al. HVAC

HVAC: Test case definition

Test case definition

LES with Smagorinsky model

Standard air at 15o C

Inlet based on a fully developed duct flow computed with RANSmodel (4 eqs v2f ) with DF-SEM for turbulent fluctuations

Bulk velocity UB = 7.5 m/s Re 40000Low-Re Reynolds stress model also tested but flow is re-laminarize

Free inlet/outlet BC at the side and at the exit of the plenum

BC based on the Bernoulli relation between the face and a point onthe same stream-line place at in case of incoming flowHomogeneous Neumann is applied on the velocityAssuming u = 0 the dynamic pressure at the face is:

pf = 1 + K

2f uf uf

with K being a head loss.

Rolfo et al. HVAC

Flow Visualisation

(Loading movie)

Figure 1: Iso-surfaces of Vorticity.

Rolfo et al. HVAC

Lavf55.48.100

HVAC_IsoSurfOmega_framerate10.mp4Media File (video/mp4)

Mean Flow (Mid Plane)

Flow separation after the bend

Impingement of the separatedflow on the obstacle

Recirculation behind the obstacle

Two counter-rotating vortices

Rolfo et al. HVAC

Mean Flow (Mid Plane)

Flow separation after the bend

Impingement of the separatedflow on the obstacle

Recirculation behind the obstacle

Two counter-rotating vortices

Rolfo et al. HVAC

Mean Flow (Mid Plane)

Flow separation after the bend

Impingement of the separatedflow on the obstacle

Recirculation behind the obstacle

Two counter-rotating vortices

Rolfo et al. HVAC

Mean Flow (Mid Plane)

Flow separation after the bend

Impingement of the separatedflow on the obstacle

Recirculation behind the obstacle

Two counter-rotating vortices

Rolfo et al. HVAC

Pressure RMS (Mid Plane)

Pressure fluctuations arepeaking in recirculation areas

These areas are candidate to bethe location of the noise sources

Rolfo et al. HVAC

Pressure RMS (Mid Plane)

Pressure fluctuations arepeaking in recirculation areas

These areas are candidate to bethe location of the noise sources

Rolfo et al. HVAC

Pressure RMS (Mid Plane)

Pressure fluctuations arepeaking in recirculation areas

These areas are candidate to bethe location of the noise sources

Rolfo et al. HVAC

3D effets

XZ plane (obstacle plane)Mid Plane 2y/DH = 0.5

2y/DH = 0.75 2y/DH = 0.85

Rolfo et al. HVAC

3D effets

YZ plane (View from bending centre)Top wall x/DH = 0.01 x/DH = 0.06

x/DH = 0.125 x/DH = 0.375

Rolfo et al. HVAC

3D effets

XY plane (Streamwise)Inlet bend = 90o = 60o

= 30o Outlet bend = 0o

Rolfo et al. HVAC

3D effets

XY plane (Streamwise)Near outlet bend z/DH = 0.125 z/DH = 1.0

z/DH = 1.75 z/DH = 2.0

Rolfo et al. HVAC

Wall pressure fluctuations

101 102 103

f [Hz]

40

50

60

70

80

90

100

110

120

SPL

[dB

/Hz]

Point @ [ 0.04 0. 0.034]

FFT

Periodogram

Welch

EXP

P1

101 102 103

f [Hz]

40

50

60

70

80

90

100

110

120

SPL

[dB

/Hz]

Point @ [ 0.04 0. 0.239]

FFT

Periodogram

Welch

EXP

P2

101 102 103

f [Hz]

40

50

60

70

80

90

100

110

120

SPL

[dB

/Hz]

Point @ [ 0.04 0. 0.277]

FFT

Periodogram

Welch

EXP

P3

Rolfo et al. HVAC

Paraview on a cluster

Why Paraview?

It is not always possible to use co-processing (i.e. Catalyst)

Possible to use remote machines with Client/Server mode

We have a visualization cluster: maybe we can use it:

2 fat nodes with 32 computing nodes and 64GB RAM2 GPUs attached to each node

The Pipeline

Load of the data EnsightReaderConvert data fromcells to nodes CellToPointGradient GradUn-structuredMeshUpdate the layout GeoRepresentation

Rolfo et al. HVAC

Scalability Pipeline in

loading a time step

100 101

no MPI tasks

101

102

Tim

e [

s]

Scalability Total Time

Linear Scaling

1 node 1 GPU

1 node 2 GPU

2 nodes Host 1

2 nodes Host 2

Rolfo et al. HVAC

Scalability individual Filters

100 101

no MPI tasks

10-1

100

101

Tim

e [

s]

Scalability Ensight Reader

Linear Scaling

1 node 1 GPU

1 node 2 GPU

2 nodes Host 1

2 nodes Host 2

100 101

no MPI tasks

10-1

100

101

Tim

e [

s]

Scalability Cell Data To Point DataLinear Scaling

1 node 1 GPU

1 node 2 GPU

2 nodes Host 1

2 nodes Host 2

100 101

no MPI tasks

100

101

Tim

e [

s]

Scalability GradientLinear Scaling

1 node 1 GPU

1 node 2 GPU

2 nodes Host 1

2 nodes Host 2

100 101

no MPI tasks

10-1

100

101

Tim

e [

s]

Scalability GeoRepresentation

Linear Scaling

1 node 1 GPU

1 node 2 GPU

2 nodes Host 1

2 nodes Host 2

Rolfo et al. HVAC

Conclusions

and Future Work

HVAC

LES of a simplified HVAC has been presented and the availableresults are in agreement with experimental data available

Application of Client/Server Paraview mode has been shown andpoor scalability of the Ensight reader has been identified

Future work

Further comparison with experimental value (hydrodynamic)

Further investigations into aeroacoustic using the Curles analogy

More calculations with different SGS models are on going

Awarding of a 1 year project to implement a FWH aeroacousticmodule into Code Saturne

Rolfo et al. HVAC

Heat and Ventilation Air Conditioning: HVACIntroductionFlow FeaturesAcoustic

Paraview in a Visualization Cluster (Placement of E. Harrison)Conclusions and Future Work